Research On Switching Monitoring Of Abnormal Consumers Based On Non-intrusive Load Decomposition

With the rapid increase in the types of electrical appliances used by users,the number of electrical accidents caused by this has increased year by year.In view of this,real-time monitoring of the use and operation of electrical appliances of power users is of great significance to better protect the safety of power users’ lives and properties.For the monitoring of users’ abnormal electrical appliances,taking into account the monitoring cost and the acceptability of the monitored users,this dissertation adopts the method of non-intrusive load decomposition to monitor the user’s specific electrical appliances,that is,only need to collect the power consumption on the user bus.Data,supported by machine learning related models,complete real-time monitoring of users’ abnormal electrical appliances.In order to ensure the accuracy and time performance of monitoring,while avoiding missed detections and false detections as much as possible,the design of the deep learning model is very important.This dissertation first conducts a more systematic research on the non-intrusive load decomposition model,summarizes the non-intrusive load decomposition research framework and the current main research methods and existing problems of each module,and focuses on the principles of the model used in subsequent research In addition,the current mainstream performance evaluation index system is constructed on the basis of further clarification and definition of the switching monitoring research to provide support for the subsequent evaluation of the model proposed in this article.The research scenarios in this dissertation are mainly divided into the following two types.One is to monitor the switching of electrical appliances other than conventional electrical appliances.Aiming at this scenario,this dissertation uses the convolutional neural networkbased Siamese neural networks to extract features based on the V-I trajectory characteristic curve,and combines the fuzzy C-means clustering algorithm to classify conventional electrical appliances and abnormal electrical appliances.This dissertation conducts experiments on the model performance based on the public data set PLAID,and verifies the effectiveness of the proposed model.The other is mainly for the switching and monitoring of some specific abnormal electrical appliances.In this scenario,this article is based on user power characteristics.In order to fully tap the timing characteristics of user power information,this article mainly uses the improved BERT based on Transformer-XL as the load.The monitoring model is used to obtain the decomposition results,and the threshold value is used to determine the switching status of abnormal electrical appliances.In this process,in order to better predict the operating status of electrical appliances,this dissertation also improves the loss function of the model accordingly.In order to verify the superiority of the model,this article is based on the public data set REDD,and compared with a number of current models with better performance.